Support vector machines using GMM supervectors for speaker verification

IEEE Signal Processing LettersPublished 12 April 2006

William M. Campbell, Douglas Sturim, D.A. Reynolds

DOI10.1109/lsp.2006.870086

Source URL

Citations1,020

SJR quartileQ1

SJR score0.94

SNIP1.40

TL;DR

This work examines the idea of using the GMM supervector in a support vector machine (SVM) classifier and proposes two new SVM kernels based on distance metrics between GMM models that produce excellent classification accuracy in a NIST speaker recognition evaluation task.

Abstract

Gaussian mixture models (GMMs) have proven extremely successful for text-independent speaker recognition. The standard training method for GMM models is to use MAP adaptation of the means of the mixture components based on speech from a target speaker. Recent methods in compensation for speaker and channel variability have proposed the idea of stacking the means of the GMM model to form a GMM mean supervector. We examine the idea of using the GMM supervector in a support vector machine (SVM) classifier. We propose two new SVM kernels based on distance metrics between GMM models. We show that these SVM kernels produce excellent classification accuracy in a NIST speaker recognition evaluation task.

Keywords

Computer Science

Digital Signal ProcessingSpeaker Verification Using Adapted Gaussian Mixture Models

4,268 Citations2000Douglas A. Reynolds, Thomas F. Quatieri +1 more

The major elements of MIT Lincoln Laboratory's Gaussian mixture model (GMM)-based speaker verification system used successfully in several NIST Speaker Recognition Evaluations (SREs) are described.

Wiley Interdisciplinary Reviews Computational StatisticsSupport vector machines

692 Citations2009Alessia Mammone, Marco Turchi +1 more

IEEE Transactions on Speech and Audio ProcessingEigenvoice modeling with sparse training data

472 Citations2005Patrick Kenny, Gilles Boulianne +1 more

This work derives an exact solution to the problem of maximum likelihood estimation of the supervector covariance matrix used in extended MAP (or EMAP) speaker adaptation and shows how it can be regarded as a new method of eigenvoice estimation.

Applied Physics Letters10.1162/15324430152733142

389 Citations2000Ronan Collobert, Samy Bengio

A decomposition algorithm, SVMTorch, similar to SVM-Light proposed by Joachims (1999) for classification problems, but adapted to regression problems is proposed, which can now efficiently solve large-scale regression problems.

View PDF

A Kullback-Leibler Divergence Based Kernel for SVM Classification in Multimedia Applications

379 Citations2003Pedro J. Moreno, Purdy Ho +1 more

This paper suggests an alternative procedure to the Fisher kernel for systematically finding kernel functions that naturally handle variable length sequence data in multimedia domains and derives a kernel distance based on the Kullback-Leibler (KL) divergence between generative models.

Support Vector Machines

373 Citations2007Jaime Gòmez Sàenz de Tejada, Juan Seijas Martìnez-Echevarrìa

A support vector machine (SVM) searches for observations that are observations that are found to lie at the edge of an area in space which presents a boundary between one of these classes of observations and another class of observations.

Advances In Channel Compensation For SVM Speaker Recognition

239 Citations2006A. Solomonoff, William M. Campbell +1 more

This work performs channel compensation in SVM modeling by removing non-speaker nuisance dimensions in the SVM expansion space via projections via an eigenvalue problem.

2003 IEEE International Conference on Acoustics, Speech, and Signal Processing, 2003. Proceedings. (ICASSP '03).Channel robust speaker verification via feature mapping

229 Citations2003D.A. Reynolds

A new feature mapping technique that maps feature vectors into a channel independent space is presented that learns mapping parameters from a set of channel-dependent models derived from a channel-independent model via MAP adaptation.

IEEE Signal Processing LettersFast approximation of Kullback-Leibler distance for dependence trees and hidden Markov models

224 Citations2003N. Minh

A fast algorithm to approximate the Kullback-Leibler distance (KLD) between two dependence tree models is presented, which offers a saving of hundreds of times in computational complexity compared to the commonly used Monte Carlo method.

IEEE Transactions on Speech and Audio ProcessingSpeaker verification using sequence discriminant support vector machines

198 Citations2005Vincent Wan, Steve Renals

This paper presents a text-independent speaker verification system using support vector machines (SVMs) with score-space kernels and introduces a technique called spherical normalization that preconditions the Hessian matrix.

View PDF